Published on 15 September 2026
Memantine in dementia: what does a small but certain effect change?
In brief
The reference Cochrane review of memantine brings together 44 trials and almost 10,000 participants. Its conclusion is clear and rarely quoted as it stands: in moderate-to-severe Alzheimer’s disease the benefit is established at high certainty, consistently across four domains, cognition, activities of daily living, behaviour and global clinical rating. It is small. In mild disease it is not demonstrated, and discontinuations for adverse events may double. What the review leaves open is not whether memantine acts, but whether what it gains is large enough to matter to the patient and to the person who looks after them.
The context
Memantine is a moderate affinity uncompetitive antagonist of glutamate NMDA receptors. It is licensed for moderate and severe Alzheimer’s disease, and in the United States it is also widely used off-label in mild disease.
The clinical question is narrower than the pharmacological one. Does memantine change anything the patient or the carer can notice, at which stage of the disease, and at what price in tolerability, in people who are old, frail and usually taking several other drugs? A review that pools every placebo-controlled trial at the licensed dose is the right instrument for that question, provided its numbers are read exactly as they are written.
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| Patients with dementia, all causes and all stages. Of 44 trials and almost 10,000 participants, 29 trials and 7,885 participants concern Alzheimer’s disease | |
| Intervention | |
| Memantine at the licensed dose only: 20 mg per day, or 28 mg extended release | |
| Comparator | |
| Placebo, double-blind parallel group randomised trials | |
| Outcomes | |
| Four clinical domains: global rating, cognitive function, activities of daily living, behaviour and mood. Analyses restricted to data at six to seven months | |
| Design | |
| Cochrane systematic review with meta-analyses, search up to 25 March 2018, certainty rated with GRADE · CEBM 1a |
One methodological point counts more here than elsewhere. The authors state it plainly: “For nearly half the studies, relevant data were obtained from unpublished sources.” The sources are trial registries, manufacturers’ press releases and posters, regulators’ websites, direct contact with authors and companies. In a field where publication bias has distorted estimates for years, that search markedly reduces the risk of seeing only the favourable trials.
The findings
Moderate-to-severe Alzheimer’s disease: high certainty, small effect
High-certainty evidence, from up to fourteen studies in around 3,700 participants, shows a small but consistent clinical benefit across the four domains. Most of the included trials are described by the authors as being “at low or unclear risk of bias”.
| Domain | Scale | Benefit over placebo (95% CI) | Certainty |
|---|---|---|---|
| Clinical global rating | CIBIC+ | 0.21 points (0.14 to 0.30) | High |
| Cognitive function | SIB | 3.11 points (2.42 to 3.92) | High |
| Activities of daily living | ADL19 | 1.09 points (0.62 to 1.64) | High |
| Behaviour and mood | NPI | 1.84 points (1.05 to 2.76) | High |
The coherence across the four domains is what gives this result its weight: this is not a single outcome that happened to cross a threshold, but an effect found everywhere, in the same direction. Discontinuations may not differ from placebo, with a risk ratio of 0.93 (0.83 to 1.04), at low certainty. Taking a cholinesterase inhibitor at the same time does not change the difference between memantine and placebo, with two possible exceptions: the effect on behaviour would be larger when the two are combined, and the effect on cognition smaller.
On agitation, a distinction has to be made. Memantine reduces the risk of agitation arising as an adverse event, with a risk ratio of 0.81 (0.66 to 0.99) at moderate certainty. Three additional studies suggest, also at moderate certainty, that “memantine is not beneficial as a treatment for agitation” once it is established. Preventing is not treating.
The other situations
| Situation | Result |
|---|---|
| Mild Alzheimer’s disease (MMSE 20 to 23) | Probably no difference in cognition, daily function or behaviour: ADAS-Cog 0.21 (−0.95 to 1.38), ADL23 −0.07 (−1.80 to 1.66), NPI −0.29 (−2.16 to 1.58). There may be no difference in global rating: CIBIC+ 0.09 (−0.12 to 0.30). Discontinuations because of adverse events may be more frequent: risk ratio 2.12 (1.03 to 4.39) |
| CertaintyMainly moderate, lower for the global rating and for discontinuations. Post-hoc subgroups, around 600 participants | |
| Mild-to-moderate vascular dementia | Probably a small cognitive benefit: ADAS-Cog 2.15 (1.05 to 3.25). There may be a small benefit on behaviour. Probably no difference in global rating, and there may be none on daily function |
| CertaintyModerate to low, 2 studies, around 750 participants | |
| Dementia with Lewy bodies and Parkinson’s disease dementia | Possible small benefit on the global rating |
| CertaintyLow to very low, 4 studies, 319 participants | |
| Frontotemporal dementia | Limited data, no conclusion possible |
| CertaintyLow to very low, 2 studies, 133 participants | |
The contrast between stages is the most directly usable piece of information. In mild disease the benefit is not merely undemonstrated: discontinuations for adverse events may double, and the lower bound of that interval sits at 1.03, which is a reason not to harden the figure. The authors put it plainly for early prescribing, writing that “at present the evidence is against this, despite it being common practice”. Prescribing at that stage means exposing the patient without a demonstrated return.
On overall tolerability the review is reassuring and precise. High certainty for two results: no difference in the proportion of patients experiencing at least one adverse event, risk ratio 1.03 (1.00 to 1.06), and no difference in falls. At lower certainty, memantine would raise the risk of dizziness about 1.6-fold, 6.1% versus 3.9%, at moderate certainty, and the risk of headache 1.3-fold, 5.5% versus 4.3%, at low certainty.
Critical appraisal
| Domain | Finding |
|---|---|
| Clinical relevance | |
| The central point: 3.11 points on a scale that runs to 100, and 1.84 points on a scale that runs to 144. Certainty is high on the existence of the effect, not on its importance to the patient and those around them | |
| Age of the evidence | |
| Search conducted up to March 2018. The review does not include work published since | |
| Origin of the trials | |
| A substantial share of the data comes from unpublished sources obtained from the manufacturers. That approach reduces publication bias without erasing the industry origin of the pivotal trials | |
| Post-hoc subgroups | |
| The results in mild disease rest on subgroups defined after the fact, which lowers their certainty | |
| Rare aetiologies | |
| Small numbers in dementia with Lewy bodies and Parkinson’s disease dementia (319 participants), in frontotemporal dementia (133) and in AIDS-related dementia complex (140): the absence of a conclusion is not a negative conclusion | |
What high certainty does not settle
The review answers one question and leaves the next one open. It establishes that an effect exists, quantifies it stage by stage and domain by domain, and states how much confidence each estimate deserves. It does not say whether those points cross the threshold at which a family notices a difference. The authors come close to saying so themselves when they write that clinical heterogeneity makes a large effect size unlikely for any single drug, and that the optimal treatment may involve several drugs, “each having an effect size that may be less than the minimum clinically important difference”.
The argument about memantine is about what the effect is worth, not about whether it exists. That distinction is a working case for evidence-based medicine: showing that an effect is real never decides anything on its own, because someone still has to judge whether it matters for this patient and whether it justifies the exposure. The same argument runs through the debate on the anti-amyloid antibodies, where a Cochrane review again finds a difference that is statistically present and, in size, hard to call meaningful.
Level of evidence
PEB appraisal: high certainty in moderate-to-severe Alzheimer’s disease, mainly moderate certainty in mild disease and in vascular dementia, low certainty in the other aetiologies. This level of evidence is uncommon in old age psychiatry. It supports the existence of a benefit that is small, consistent and reproducible, not the claim that it is decisive for the patient’s trajectory.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ High certainty across four domains is rare and it is not up for debate. What is up for debate is three SIB points out of a hundred: will the family see them? I prescribe it in moderate to severe disease if it is well tolerated, not in mild disease, and I say honestly that the gain is a small one. ”
What this means in practice: the argument is not about whether the drug acts, it is about whether the effect is worth having for this particular patient. That is a conversation, not a unilateral decision.
What you can do with this on Monday morning.
- In moderate-to-severe Alzheimer’s disease, memantine remains defensible when tolerability is good, provided the real size of the expected benefit is stated before the first prescription.
- In mild disease, do not prescribe: no demonstrated benefit, and twice as many discontinuations for adverse events.
- Faced with agitation that is already established, do not expect a therapeutic effect from it. The review separates preventing agitation from treating it.
- Do not stop memantine on the grounds that the patient is already taking a cholinesterase inhibitor: the benefit is found either way, and the plain language summary states that adding memantine to established treatment also results in less deterioration than placebo.
And when the family asks whether it is worth it, which happens at almost every consultation, give them the measurement rather than a verdict: the effect is real, it has been quantified, it is small, and nothing here shows that it changes the course of the disease. Then set a review date, keep the drug only while it is tolerated, and say in advance what would make you stop. That takes longer than saying the drug does not work, and it is more faithful to what has been measured.
Frequently asked questions
Is memantine licensed at every stage of Alzheimer’s disease?
No. The licence covers moderate and severe disease. In the United States the drug is also widely used off-label in mild disease, and that is precisely the situation in which this review finds no demonstrated benefit and a possible doubling of discontinuations for adverse events. Prescribing at that stage is a choice made outside the licence and outside the evidence.
Is a gain of three points on a hundred-point scale clinically useful?
This is the question the review does not settle. It measures the effect and grades the confidence attached to it, it does not set the threshold at which the effect becomes worth having. The authors note that no single drug is likely to produce a large effect size in this disease, and that useful treatment may end up combining several drugs whose individual effects fall below the minimum clinically important difference. The answer depends on the stage, on what the family expects and on individual tolerability. It belongs to a shared decision, not to a universal threshold.
Should memantine be combined with a cholinesterase inhibitor?
The benefit of memantine is found whether or not the patient is taking a cholinesterase inhibitor, and the review reports that adding memantine to established treatment also results in less deterioration than placebo. Two possible qualifications are flagged: a larger effect on behaviour when the two are combined, and a smaller effect on cognition.
What can be said about frontotemporal dementia?
Two trials and 133 participants, at low to very low certainty. There is no demonstration of efficacy, and the review reports no tolerability data specific to this aetiology. Where no benefit is established, doing nothing remains the easiest position to defend.
And in vascular dementia?
A small cognitive benefit is probable, about 2 ADAS-Cog points, with no demonstrated effect on daily function or on the global rating. That is little, and it rests on two trials only.
Annotated bibliography
McShane R, Westby MJ, Roberts E, Minakaran N, Schneider L, Farrimond LE, Maayan N, Ware J, Debarros J (2019). Memantine for dementia. Cochrane Database of Systematic Reviews, 3, CD003154. DOI 10.1002/14651858.CD003154.pub6 · PMID 30891742. Sixth edition of the review. Source study analysed here.